Enhanced EEG Classification Using Gradient Boosting and Volterra Kernel-Based Feature Extraction

Electroencephalography (EEG) signal classification presents important challenges due to signal complexity, noise, and class variance. This study introduces an advanced preprocessing and feature extraction pipeline that utilizes signal detrending, bandpass filtering, and segmentation for improved data quality. A diverse set of features including pseudo- covariance, IEK covariance, wavelet transformations, spectral entropy, and Volterra kernel estimation capture the intricate properties of EEG signals. Different classifiers inclusive of Support Vector Machines (SVM), Multi-Layer Perceptron (MLP), Gradient Boosting, and K-Nearest Neighbours (KNN) are validated using accuracy, precision, recall, and F1-score metrics. The proposed binary classification model demonstrates the gradient boosting classifier achieves an accuracy of 99.5% across all the EEG class comparisons, SVM (Radial basis function Kernel) indicated strong performance, ranging from 92.6% to 97.0% accuracy, while KNN achieved up to 98.8% accuracy. In compared to, MLP yielded lower accuracy values (between 51.1% and 53.1%), indicating limitations in capturing EEG signal complexities. Experimental results from Bonn University’s EEG dataset demonstrate the efficacy of these techniques in distinguishing EEG classes, paving the way for more robust biomedical signal processing methods. This research contributes to the advancement of EEG classification for potential applications in healthcare diagnostics and brain-computer interface development.